Page 17 - FIGI - Big data, machine learning, consumer protection and privacy

• Social media data decisions (or to provide inputs to human deci-
• B2B data acquired from parties in the supply chain sions). Automated decisions are decisions made by
• Agriculture data (e.g., feeds on corn production) computer processing systems without any human
• Point of sale data involvement (beyond the coding), typically based
• Pharmaceutical prescription data on inferences produced by profiling using machine
learning models applied to big data. Inferences and
The increasing connectivity of devices provides predictions improve firms’ ability to discriminate
opportunities for data for financial services provid- among consumers, offering them products and ser-
ers. For instance, cars today have extensive comput- vices suited to their preferences or needs, and at
ing power, use extensive code, and process huge prices they are willing to pay. Examples include deci-
amounts of data. Lenders increasingly require sions whether to extend credit to an individual or to
35
borrowers, particularly higher risk (subprime) offer the person a job.
borrowers, to consent to installation of starter-in- Numerous applications of big data and machine
terrupter devices (SIDs) or other tracking devices in learning are being introduced in financial services,
their cars when providing a loan. SIDs have the prac- including:
tical benefit of supporting enforcement of reposses-
sion rights by enabling the lender to disable a vehicle • risk assessment, whether for lending or insurance,
if the borrower defaults on the loan. At the same time, as discussed above, by companies such as Com-
they and other tracking devices supply data such pare.com; 38
as daily driving activities and locations which allow • investment portfolio management “robo-advis-
inferences about home and work addresses, whether ers” such as Betterment and Wealthfront that
40
39
the person is still driving to a regular place of employ- rely on algorithms to calibrate a financial portfolio
ment (and so employment status), where the person to a consumer’s investment goals and tolerance
likes to shop or be entertained, and departures from for risk;
habits that may indicate changes in preferences. • high-frequency trading (HFT) by hedge funds and
Tracking devices may also supply data about driving other financial institutions such as Walnut Algo-
behaviour patterns that indicate not only skill levels rithms and Renaissance Technologies that use
42
41
but sometimes even a particular emotional or mental machine learning for making trading decisions in
state (e.g., repeatedly accelerating unusually quickly, real time; 43
or breaking unusually abruptly). • asset management, liquidity and foreign currency
Today, a substantial market in inferences about risk and stress testing;
people now exists, and how these are generated and • fraud detection by companies like APEX Ana-
used is discussed in the next section. Overall, the lytics and Kount through detection and flag-
45
44
relation between artificial intelligence and big data ging of unique activities or behaviour anomalies
is “bi-directional.” Big data relies on artificial intelli- to block transactions and for security teams to
gence and machine learning to extract value from investigate; and
big datasets, and machine learning depends on the • a host of services such as security and digital
vast volume of data in order to learn. 36 identification, news analysis, customer sales and
recommendations, and customer service. 46
2�3 What are profiling and automated decisions?
Big data, machine learning and artificial intelligence In some cases, these new uses are supported by
(AI) are enabling profitable commercial opportuni- legislation expressly authorising the use of artificial
ties and social benefits through profiling and auto- intelligence. For instance, Mexico’s fintech reforms in
mated decisions. 2018 amended the Securities Market Law to allow for
Profiling is the automated processing of personal special rules for automated advisory and investment
data to evaluate, analyze or predict likely aspects of management services (also known as robo-advis-
a person's interests, personal preferences, behaviour, ers). 47
performance at work, economic situation, health,
reliability, location or movements. Data analytics 2�4 What is consumer protection?
37
enables the identification of links between individu- Consumer protection is designed to protect humans
als and the construction of group profiles. where they are vulnerable. These may include
Such inferences and predictions may be used protection of children, the elderly, and others who
for targeted advertising, or to make automated cannot protect themselves for physical or psycho-

Big data, machine learning, consumer protection and privacy 15